Adaptive threshold from the Sentinel-2 oil spill index for oil spill detection

Abstract

One method for detecting oil spills from optical satellite data is using Oil Spill Index (OSI). In several studies, this index was used to visualize oil spill distribution. In order to estimate the coverage area of oil spill that occurs, image segmentation needs to be carried out to separate object of oil spill from non-oil spills. Therefore, this study used comparison of several OSI index applied to Sentinel-2 images and segmentation of oil and non-oil objects from OSI images using index threshold values. The data used consist of Sentinel-2 data date 12 August 2021 in Karawang waters, and 5 October 2019 and 2 May 2021 in Bintan waters. Karawang water image indicated oil spill from oil well platform leakage and Bintan waters image showed oil spills from ship sewage and other unknown sources. Two OSI algorithms were used, that is OSI1 that had been developed using Sentinel-2 image, and OSI2 was developed using MODIS data. OSI image threshold was then used to separate oil and non-oil objects. Other band combinations thresholds were also used to obtain the better results. Based on the result of threshold on OSI1 and OSI2 images, the study found that oil object can be separated better in OSI2 image, however in the two images there were still a lot of non-oil subject mixed with oil class especially in OSI1. Based on analysis of spectral pattern, object separation in OSI1 image need to be carried out further using band 2 and band 9 threshold, while for OSI2 needs to be separated further using band 3 threshold. Based on the comparison of the 2 thresholds, using combination of OSI2 and band 3 gave better result. Accuracy analysis of OSI2 threshold also showed the better result with overall accuracy of 86%.